Full-component reconstruction of three-dimensional fluid stress tensors

Shunsuke Kumagai, Shun Miyatake, Ryusuke Cho, William Kai Alexander Worby, Masanori Naito, Takahiro Ushioku, Masanobu Horie, Yoshiyuki Tagawa · 2026

Read Full-component reconstruction of three-dimensional fluid stress tensors by Shunsuke Kumagai, Shun Miyatake, Ryusuke Cho, William Kai Alexander Worby, Masanori Naito, Takahiro Ushioku, Masanobu Horie, Yoshiyuki Tagawa (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23309.txt.

Introduction

Kumagai and seven coauthors present U-FlowPET, a self-taught physics-informed setup joining light-stress scans with the ruling laws of fluid flows. It rebuilds the full 3D stress tensor without fixed guesses or tagged training data. Rather than learning from labeled known stress fields, the method finds nature-valid stress fields meeting motion balance and flow while staying matching with seen optical views. In round-pipe flow with an exact answer, all six stress parts are rebuilt with scaled mean absolute errors below 4 percent.

Catalog record

CREATORShunsuke Kumagai, Shun Miyatake, Ryusuke Cho, William Kai Alexander Worby, Masanori Naito, Takahiro Ushioku, Masanobu Horie, Yoshiyuki Tagawa
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50708
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.23309

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Full-component reconstruction of three-dimensional fluid stress tensors Shunsuke Kumagai, Shun Miyatake, Ryusuke Cho, William Kai Alexander Worby, Masanori Naito, Takahiro Ushioku, Masanobu Horie, Yoshiyuki Tagawa Abstract Forces govern how fluids deform biological tissues, regulate cardiovascular function, and determine the performance and failure of soft materials. Recent advances in flow birefringence, including the use of suspended anisotropic nanomaterials to optically encode stress in fluids, have made direct stress measurement experimentally accessible in projection. However, direct experimental access to all six components of the three-dimensional (3D) fluid stress tensor has remained unattainable because optical measurements provide only path-integrated observables. Recovering local 3D stresses from such data constitutes an intrinsically underdetermined tensor tomography problem, in which two optical observables must determine six independent stress components. Here we introduce U-FlowPET, an unsupervised physics-informed framework that integrates photoelastic tomography with the governing equations of fluid mechanics to reconstruct the full 3D stress tensor without relying on constitutive assumptions, geometric symmetry, or labeled training data. Rather than learning from labeled reference stress fields, the method identifies physically admissible stress fields that satisfy momentum balance and continuity while remaining consistent with measured optical projections. We validate the approach using analytical, numerical, and experimental datasets. In axisymmetric pipe flow with an analytical solution, all six stress components are reconstructed with normalized mean absolute errors below 4%. Robust reconstruction is further demonstrated in geometrically complex curved-pipe flow without symmetry assumptions and in experimental pipe-flow data despite measurement noise. By enabling direct 3D stress-field reconstruction from optical data alone, U-FlowPET extends fluid analysis from observing motion to quantifying force and establishes a new framework for stress-based diagnostics in biological flows and functional materials. \acknow This work was supported by JSPS KAKENHI Grant No. JP24H00289 (Y.T.) and JST PRESTO Grant No. JPMJPR21O5 (Y.T.), Japan. † † lead-author: Kumagai † † author-contributions: Y.T. conceived and supervised the study. Y.T. and M.H. designed the research. S.K., S.M., R.C., and W.K.A.W. built the experimental system and performed the experiments. S.K., M.N., T.U., and M.H. developed the machine-learning methodology. W.K.A.W. contributed to the theoretical framework of photoelastic stress measurement. S.K. analyzed the data. S.K. and Y.T. wrote the paper with input from W.K.A.W. and T.U. All authors discussed the results and edited the manuscript. Y.T. acquired funding. † † author-declaration: The authors declare no competing interests. † † corresponding-author: * To whom correspondence should be addressed. E-mail: [email protected] \dropcap For over a century, fluid mechanics has relied primarily on velocity fields to understand flow-driven phenomena. From particle image velocimetry (PIV) 9 ; 8 ; 48 ; 49 ; 51 to modern three-dimensional diagnostics, the motion of the tracer in a fluid flow has been measured to infer deformation and transport. However, velocity is a kinematic quantity—the result of forces, not their cause. In biological systems, vascular remodeling and cellular mechanotransduction respond to stress rather than velocity 17 ; 18 ; 22 ; 37 ; 58 ; 57 ; 56 , and in soft robotics and functional materials, mechanical performance is governed by internal force transmission 12 ; 60 ; 35 ; 36 ; 47 . Direct experimental access to the full three-dimensional (3D) stress tensor field in fluids would therefore mark a transition from kinematic observation to dynamic understanding. Despite its importance, obtaining all six independent components of the 3D fluid stress tensor remains experimentally challenging. The Cauchy stress tensor in three dimensions is represented by a 3\times 3 matrix (nine components), which reduces to six independent components due to symmetry. Conventional pressure sensors 34 ; 13 provide only localized wall measurements and cannot resolve internal stress structures non-invasively. Velocity-based methods indirectly infer stresses: The strain rate tensor is computed by spatial differentiation of measured velocities, and stresses are estimated through a constitutive relation linking strain rate to stress. This procedure amplifies measurement noise and becomes unreliable in flows with unknown or complex constitutive behavior, including non-Newtonian fluids. Photoelasticity 39 ; 5 ; 2 ; 21 ; 28 ; 20 offers a fundamentally different approach by directly encoding stress-induced anisotropy into optical birefringence without requiring spatial differentiation. However, the measured photoelastic parameters—retardation \Delta and principal orientation \phi —are path-integrated projections of stress along the optical axis, and recovering local 3D stresses from these projections requires tomography. Unlike scalar tomography, which reconstructs quantities such as density 50 ; 16 ; 23 ; 45 , photoelastic reconstruction constitutes a tensor tomography problem: two optical observables must determine six independent components of a 3D tensor field, rendering the inverse problem intrinsically under-determined. Historically, this information deficiency confined reconstruction to simplified cases. Classical integrated photoelasticity succeeded mainly under axisymmetric or two-dimensional assumptions, where strong geometric symmetry reduces the number of unknowns 6 ; 1 ; 4 ; 19 ; 15 ; 30 ; 32 . Extensions to arbitrary geometries proved unstable under limited viewing angles and experimental noise 24 ; 33 ; 54 ; 7 ; 10 ; 3 . Analytical inversion collapses without imposing constitutive relations, symmetry, or restrictive regularization. As a result, full 3D stress tensor reconstruction in flowing fluids has remained unresolved. Recent advances in inverse problems and machine learning have enabled recovery of high-dimensional fields from incomplete data 43 ; 44 ; 14 ; 29 . Convolutional neural networks and physics-informed frameworks show promise, yet existing stress-reconstruction approaches rely on supervised learning with ground-truth stress fields that are experimentally unavailable 27 ; 52 ; 55 . The central question therefore remains: Can this information-deficient tensor tomography problem be closed without constitutive models, geometric symmetry, or labeled training data? Figure 1: Concept of U-FlowPET for full-component reconstruction of three-dimensional fluid stress tensors. Auxiliary solid and dashed black lines in the fluid stress tensor fields are used to highlight boundaries and cross-sectional contours. Multi-angle photoelastic measurements provide two optical observables: retardation \Delta and principal orientation \phi , as path-integrated projections of stress in a flowing birefringent fluid ( Movie S1 ). U-FlowPET integrates optical data consistency with governing fluid mechanics constraints to reconstruct the full three-dimensional Cauchy stress tensor field ( 3\times 3 tensor with six independent components). The framework operates in an unsupervised manner, without constitutive assumptions, geometric symmetry, or labeled training data, and enables symmetry-free and noise-robust reconstruction of internal stress fields. Here, we introduce U-FlowPET (Unsupervised flow-integrated photoelastic tomography), a physics-informed unsupervised framework that integrates photoelastic tomography with the governing equations of fluid mechanics to reconstruct full 3D stress tensor fields. Figure 1 illustrates the conceptual framework of U-FlowPET. Multi-angle measurements of retardation \Delta and principal orientation \phi provide two optical observables that represent path-integrated projections of stress. By enforcing consistency with these optical data while simultaneously satisfying momentum balance and the continuity equation of incompressible fluids, the framework reconstructs all six independent components of the three-dimensional Cauchy stress tensor ( \sigma_{xx},\sigma_{xy},\sigma_{xz},\sigma_{yy},\sigma_{yz},\sigma_{zz} ) without relying on constitutive relations or geometric symmetry assumptions. We validate the method via three test cases: synthetic axisymmetric pipe flow with analytical solutions, geometrically complex curved-pipe flows without symmetry assumptions, and experimental pipe flow data in the presence of measurement noise. Across these validations, the six stress components are reconstructed with normalized mean absolute errors below 4% in the analytical pipe-flow benchmark, within 10%–18% in the curved-pipe case, and below 8% for experimentally acquired pipe-flow data, demonstrating that physically constrained

Read the complete text